The field of time series analysis is shifting towards the integration of multimodal data, particularly textual information, to improve forecasting accuracy. This trend is driven by the recognition that traditional numerical series models are limited by their lack of contextual information. Recent innovations have focused on developing models that can effectively combine historical and predictive textual information with numerical time series data, leveraging advanced multimodal comprehension capabilities. Notable developments include the use of large language models and cross-modality alignment techniques to enhance the accuracy and efficiency of time series forecasting.

Some noteworthy papers include: Dual-Forecaster, which proposes a pioneering multimodal time series model that integrates descriptive and predictive textual information. Efficient Multivariate Time Series Forecasting via Calibrated Language Models with Privileged Knowledge Distillation, which introduces an efficient framework for multivariate time series forecasting using calibrated language models and privileged knowledge distillation.